Let's minimise tokenmaxxing: India's next AI breakthrough will come from efficiency, not bigger models

Last week, Palantir CEO Alex Karp went into overdrive and criticised AI tech giants OpenAI and Anthropic's token-based pricing models, stating they charge businesses exorbitant costs for raw computational data without delivering concrete business value.

In this context, as AI begins to deliver business impact, success for Indian organisations hinges not only on access to advanced AI models but also on their ability to deploy AI efficiently and sustainably at scale.

Indian enterprises typically operate at enormous scale, while remaining highly cost-conscious. From banks and retailers to telecom providers and public digital platforms serving millions of users, even marginal improvements in AI efficiency can deliver savings, while inefficient AI usage can be expensive. If India hopes to take pole position in the AI race, optimising token usage efficiency could be a key success factor.


GoI has positioned AI as a strategic pillar of its economic and digital future. The ambition is to democratise AI while establishing India as a global innovation hub. But while expanding access is critical, higher AI consumption is often equated with greater AI maturity. That assumption is increasingly being challenged, since 'tokenmaxxing' - whereby maximising token consumption is considered a proxy for AI adoption - is flawed.

The real challenge is that most organisations lack mature frameworks to monitor how AI is used within the organisation. Without that visibility, AI workloads often default to the most powerful, and consequently, most expensive models, even when simpler alternatives would suffice.

For instance, not every task requires a frontier model. Simpler workflows can often be handled through traditional automation, rules-based systems, smaller models or retrieval-based approaches. Likewise, techniques such as shortening prompts, caching frequently requested responses or retrieving enterprise knowledge before generating new content can reduce token consumption without compromising user experience. Complex reasoning gets directed to advanced models where the incremental value justifies the additional cost.

At enterprise scale, these small optimisation decisions are multiplied across millions of AI interactions. The result is not only lower infrastructure costs but significantly better returns on AI investment.

India's cloud transformation demonstrated that scale alone does not create efficiency. Organisations learnt to optimise workloads, monitor utilisation and control cloud spending because every unnecessary compute cycle carried a cost. AI requires the same mindset, except the resource being optimised is intelligence.

But AI also demands a new level of operational discipline. Organisations need an effective AI governance framework to maximise RoI. Guard rails must encompass not just operational oversight and policy enforcement but also economic accountability. The goal is sustainable AI economics that supports innovation to scale. Achieving this requires visibility into AI consumption across the enterprise.

As AI deployments grow, organisations need answers to operational questions. Which teams are consuming the most AI resources? Which models are delivering the greatest business value? Are simple requests being routed to expensive frontier models unnecessarily? Are AI agents complying with organisational policies?

An operational layer or control plane can help organisations monitor AI usage across the enterprise, match the right model to the right workload, enforce governance policies, track costs in real time, and optimise AI consumption. This enables enterprises to pay for the intelligence they need rather than defaulting to the highest-cost option for every task.

Ultimately, organisations that scale AI successfully will combine a trusted data foundation with operational visibility by creating an agentic control plane that connects data, context, decision-making and action in a governed and measurable way.

India's AI landscape makes this discipline especially important. The country is building AI for multilingual users, high-volume digital services, cost-sensitive businesses and citizen-scale public platforms. In such environments, success will not come from deploying the largest models everywhere, but from deploying the right level of intelligence for each interaction while maintaining affordability and trust.

While today's AI conversation often focuses on model capabilities, change is underway. The country's digital transformation has demonstrated that scale and governance are not mutually exclusive. Initiatives such as digital identity, digital payments and DPI have shown that technology can be deployed responsibly at scale while being frugal.

India's AI opportunity will not be determined by how many models it deploys, but by how efficiently and intelligently it converts every token into measurable business value. The next phase of AI leadership will belong to the organisations that treat intelligence as a finite resource that must be governed, optimised and deployed with discipline at enterprise scale.

The writer is MD, India, Snowflake